projects | Alejandro Escontrela

Gauss Gym

An open-source real-to-sim framework for learning locomotion from pixels

Holosoma

Comprehensive humanoid robotics framework for training and deploying reinforcement learning policies on humanoid robots, as well as motion retargeting. Supports locomotion (velocity tracking) and whole-body tracking tasks across multiple simulators (IsaacGym, IsaacSim, MJWarp, MuJoCo) with algorithms like PPO and FastSAC

Neural USD: Scalable Scene Editing via Differentiable Universal Scene Description

Interactive 3D scene editing using Neural USD representations

Barkour: Benchmarking animal-level agility with quadruped robots

We introduce the Barkour agility benchmark for quadruped robots, along with a Transformer-based generalist locomotion policy

DayDreamer: World Models for Physical Robot Learning

Hyper data-efficient robot learning with world models

Video Prediction Models as Rewards for Reinforcement Learning

Using advances in generative modeling to learn reward functions from unlabeled videos.

Adversarial Motion Priors Make Good Substitutes for Complex Reward Functions

Using motion priors to learn energy-efficient and natural robot locomotion strategies

Learning a Diffusion Model Policy from Rewards via Q-Score Matching

Learning diffusion model policies for off-policy reinforcement learning.

Learning to Walk on Complex Terrains with Vision

Examining the role of vision in robot locomotion

Autonomously Untangling Long Cables

Sliding and Grasping for Tangle Manipulation (SGTM)